Multi-actor outcome forecasting Influence optimization

Every decision that matters is made by a room.

The Bloc reads the public record, models everyone in the room, and forecasts where the room lands — and which actor, moved how far, lands it somewhere else.

See how it works →
01 — Method

The solver has been
public for forty years.
The inputs were the
bottleneck.

Spatial bargaining models have forecast committee, cabinet and regulatory outcomes since the early eighties. They never left the agencies and a handful of consultancies, because every issue needed an expert analyst to score the actors by hand. That is the part that got solved recently, and it is the part we build.

STEP 01

Acquire

Find and pull the record that carries the signal — filings, committee documents, transcripts, position papers, registers, attendance and amendment logs. Knowing where the signal lives is most of the work.

STEP 02

Extract

Score each actor on three parameters: position on the issue, salience — how much they care, and clout — how much weight they carry. Position is stated. Salience and clout are behavioural, and inferred from what actors spend effort on.

STEP 03

Solve

Run the bargaining game to equilibrium. Return a distribution over outcomes, not a point, and the coalition structure that produced it.

02 — Output

The forecast is not
the product. The
lever is.

Knowing where a decision lands is worth something. Knowing which actor to move, how far, and what it changes is worth considerably more.

  • Outcome distributionWhere the decision resolves, with a credible interval rather than a confident sentence.
  • Coalition mapWho ends up with whom, and which blocs form on the way to the result.
  • Counterfactual leversPer actor: how far their position or salience has to shift before the outcome moves, and by how much.
  • Source traceEvery parameter links back to the documents it was drawn from. You can audit the input, not just the answer.
  • Prediction recordCalls on live decisions, timestamped and published before they resolve. Falsifiable by anyone with a browser.
03 — Team

Four founders.
Four timezones.

Dallas

Vineel

Solver & markets

Expected utility, equilibrium solving and numerical methods. Built and traded systematic models, and knows the event-driven desk that buys this first from the inside.

Amsterdam

Soumen

Extraction & inference

LLM extraction, calibrated inference and attribution, with a publication record behind it. Builds agentic retrieval over legal and regulatory proceedings.

San Francisco

Mrunal

Forecasting & validation

Multi-horizon forecasting and self-supervised representation learning, pretrained from scratch, with the evaluation discipline underneath it. Builds the harness that tells us when we are wrong.

Bengaluru

Pawan

Acquisition layer

Source discovery and crawling at scale — millions of sites and channels evaluated for activity and relevance. Finding where salience is observable is research, not plumbing.

04 — Contact

Get in touch.

Need a decision modeled? Investing? Something else? Let us know.